Track reconstruction for the Mu3e experiment Alexandr Kozlinskiy (Mainz, KPH) for the Mu3e collaboration DPG Würzburg (.03.22, T85.

Transcription

1 Track reconstruction for the Mu3e experiment Alexandr Kozlinskiy (Mainz, KPH) for the Mu3e collaboration DPG Würzburg (.03.22, T85.1)

2 Mu3e Experiment Mu3e Experiment: Search for Lepton Flavor Violation (LFV) Decay: μ + e + e + e Standard Model: Br < (unobservable) Enhanced in New Physics models: SUSY, leptoquarks, etc. Any observed decay will point to NP SM μ + e + e + e Location: Paul Scherrer Institute (PSI) Commission in 2019, start in 2020 Current experimental status: SINDRUM (1988) Nucl.Phys.B299(1988)1 Br(μ + e + e + e ) < at 90% c.l. Neutrino mixing Mu3e aims for sensitivity of one in µ-decays Existing beam line: 10 8 µ/s With new beam line: one in Alexandr Kozlinskiy 2

3 Signal Signal: µ + e + e + e Three tracks Decay at rest Common vertex Same time p e = 0 E e = m µ e ± energy < 53 MeV/c Background: Random combinations: µ + e + + 2ν e + /e scattering Fake tracks Not same vertex, time, etc. Good vertex/time resolution Internal conversion: µ + e + e + e + 2ν Missing momentum & energy: Need good momentum resolution Alexandr Kozlinskiy 3

4 Mu3e Detector He B = 1 T O(10 8 ) µ + /s p µ ~ 28 MeV/c 12 cm r 1 ~ 23 mm r 2 ~ 29 mm Muons stop and decay on target: Double cone hollow target O(100) µm thickness Vertex separation Existing beam line at PSI: Continuous muon beam O(10 8 ) µ + /s Inner pixel layers: High granularity Thin (to reduce MS) & efficient Silicon pixel sensors (HV-MAPS) As close as possible to target Pointing to vertex Reduce effect of MS Alexandr Kozlinskiy 4

5 Mu3e Detector He B = 1 T 36 cm r 3 ~ 70 mm r 4 ~ 82 mm Fibre detector: σ t < 1 ns Suppress accidental BG Charge ID Two outer pixel layers: B = 1 Tesla Minimum p T ~ 12 MeV/c Limited by outer layer radius Alexandr Kozlinskiy 5

6 Mu3e Detector Particles (electrons) bend back in magnetic field: Use recurl stations to detect them Improve momentum resolution Factor 5-10 improvement Recurl stations: Two pixel layers (same as central station) Tile detector σ t < 100 ps Suppress accidentals constraint on radius: pixel size (min σ p ) Alexandr Kozlinskiy 6

7 HV-MAPS High Voltage Monolithic Active Pixel Sensor: Commercially available technology Large area (2 2 cm 2 ) High granularity (pixel ~ µm 2 ) Thin (~ 50 µm) Fast charge collection via drift (HV, σ t ~ 15ns) High efficiency ( > 99% ) I. Peric, Nucl.Instrum.Meth. A582 (2007) 876 Alexandr Kozlinskiy 7

8 Triplet fit Track reconstruction in mag.field Helical trajectory Require minimum 3 hits Assuming no pixel uncertainty and no energy loss (Mu3e conditions): One parameter to describe trajectory curvature r (or momentum) Triplet trajectory with multiplet scattering (MS) in middle hit MS angles are function of curvature φ ms (r), λ ms (r) Fit minimize χ 2 (scattering angle): χ 2 = φ 2 ms/σ 2 ms,φ + λ 2 ms/σ 2 ms,λ No analytical solution Small MS angles linearization around known solution (circle in xy-plane) Nucl.Instrum.Meth. A844 (2017) 135 Alexandr Kozlinskiy 8

9 Track fit Track/Segment: Sequence of triplets 3D radius: Minimize combined χ 2 Simple solution: where r i individual triplet solutions (weighted average) Note: Theoretically individual triplets can be fitted in parallel and then combined. In practice start from seed triplet and then add more hits. Alexandr Kozlinskiy 9

10 Triplets Make triplets: Combine hits of first 3 layers Hits Fibres Target combine hits n number of hits per layer Difficulty: O(n 3 ) combinations 10 hits per layer in 50 ns O(1K) combinations per frame per second large Reduce number of fits Geometrical selections (opening angles, etc) Alexandr Kozlinskiy 10

11 Triplets Fake triplets (red) Selections: Geometrical Distance between hits, opening angles, etc. Factor 50 reduction in number of fitted combinations 10 9 fits per second GPU filter farm Reduce background: triplet χ 2 Cut on MS angles Hits Fake rate (fake combinations per one truth track) ~ 4 10 truth triplets & 40 fakes Truth (MC) triplets (black) Alexandr Kozlinskiy 11

12 Short tracks short track: pair of triplets (4 hits) Make short tracks: Use triplets as seeds Estimate hit at last layer Lookup in φ/z window Combine 4 hits (triplet + hit) 2 triplets (2 shared hits) Fit (weighted average) Reconstruction frame (50 ns): O(10) short tracks Fake rate ~ 1.9% layer 4 layer 2 layer 3 layer 1 Alexandr Kozlinskiy 12

13 Long 8-hit tracks long 8-hit track (2 short tracks) Long (6- and 8- hit) tracks: Combine short track and pair of hits or two short tracks: Fake rate ~ 10-30% 1% true random combinations Rest hits from same tracks, different turns Fibre hits (one per short segment) Reject wrong combinations Charge ID layer 4 long 6-hit tracks (short + 2 hits) layer 2 layer 1 layer 3 Alexandr Kozlinskiy 13

14 Acceptance & Efficiency Full Geant4 simulation of Mu3e detector 50 ns event (frame) Decay: μ + e + υυ, 10 8 stopped muons per second ~ 5 decays within frame Acceptance (geometry): Require minimum 4 hits (1 per layer) p T inner/outer layer radius polar angle (λ) length of detector acceptance ~ 80% Reconstruction efficiency: Short tracks ~ 95% Geometrical cuts and χ 2 cuts Long tracks Used for signal reconstruction vertex fit, etc. 80% of short tracks are reconstructed as long Alexandr Kozlinskiy 14

15 Long tracks reconstruction efficiency Efficiency (acceptance) Ends of recurl stations min p T ~ 10 MeV polar angle Service areas between stations Alexandr Kozlinskiy 15

16 Momentum resolution Short tracks (4 hits) <σ p > 1.4 MeV/c Depends linearly on momentum Long tracks (6 and 8 hits) <σ p > 0.2 MeV/c min σ p 100 KeV/c Momentum resolution Long tracks σ p 0.2 MeV/c Long tracks mom.resolution Short tracks σ p 1.4 MeV/c min σ p 100 KeV/c (p min depends on λ) Alexandr Kozlinskiy 16

17 Vertex fit Signal 3 tracks (e + e + e ): Long (recurl) tracks and/or short tracks MS in first layer Pixel size & energy loss Energy loss in target Vertex: Constrained to target area Or target surface Material (first layer & target): Scattering, pixel size, energy loss Same time at vertex (fibres and/or tiles) Vertex resolution: σ z = 230 μm (limited by MS) σ x,y = 350 μm (MS + σ p ) Alexandr Kozlinskiy 17

18 Signal sensitivity Phase I detector: Main background: Radiative decay (momentum resolution) Bhabha + Michel (vertex resolution) Sensitivity: muon stops, one year of data taking Br ~ at 95 c.l. Alexandr Kozlinskiy 18

19 Summary Mu3e experiment: Search for LFV μ + e + e + e, Br < 10 15(16) Reconstruction: Use triplet fit for track reconstruction Fast, will be used offline and online (GPU filter farm) Good performance Require good momentum, space and time resolution & efficiency Short tracks: <σ p > 1.4 MeV/c Long tracks: <σ p > 0.2 MeV/c Fibre and tile time information Already meet/exceed Phase I requirements. Alexandr Kozlinskiy 19

20 Backup Alexandr Kozlinskiy 20

21 Backgrounds Internal conversion: µ + e + e + e + 2ν Missing momentum & energy: p e 0 E e m µ Need good momentum resolution Alexandr Kozlinskiy 21

22 Mu3e Detector 10 cm O(10 8 ) µ + /s p µ ~ 28 MeV/c 18 mm Muons stop and decay at rest on target: Existing beam line at PSI: Continuous muon beam O(10 8 ) µ + /s Double cone hollow target Vertex separation in space 36 mm diameter (to cover beam x-section) O(100) µm thickness Alexandr Kozlinskiy 22

23 Pixel layers Mu3e pixel layers: 2844 sensors (area ~ 1 m 2 ) sensor size 2x2 cm 2 pixel size 80x80 μm 2 50 μm thick ~ X 0 Total thickness (with support) ~ X 0 Typical MS angles (p < 53 MeV/c) ~ 5-10 mrad Sensors (50 μm, Si) Readout foil (25-50 μm, Kapton/Al) Kapton frame (25 μm) Alexandr Kozlinskiy 23

24 Triplet fit "Minimum" track in mag.field Three measurements or hits (i.e. in 3 sensor layers) Or two helices Helix trajectory defined by: Pair of hits (at the end of this helix) And curvature r (or momentum) Triplet: No hit uncertainty & MS at middle hit No energy loss (r = r 1 = r 2 ) MS angles: φ ms (r), λ ms (r) Fit minimize χ 2 (scattering angle): χ 2 = φ 2 ms/σ 2 ms,φ + λ 2 ms/σ 2 ms,λ There is no analytical solution Assume small MS angles Start from "circular" solution in xy-plane and linearize Nucl.Instrum.Meth. A844 (2017) 135 Alexandr Kozlinskiy 24

25 Triplet fit No energy loss (r = r 1 = r 2 ) No hit position uncertainty Consider only MS at middle hit Triplet: 3 hits (3D points) form triplet Combination of 2 helices Fully defined by radius r MS angles: φ ms (r), λ ms (r) Minimize χ 2 (scattering angle): χ 2 = φ 2 ms/σ 2 ms,φ + λ 2 ms/σ 2 ms,λ There is no analytical solution Assume small MS angles Start from known "circular" trajectory in xy-plane Linearization Nucl.Instrum.Meth. A844 (2017) 135 Hits Transverse view Scattering at middle hit Projection to helix "cylinder" Trajectory Alexandr Kozlinskiy 25

26 Triplet vs Karimäki vs GBL Fit performance depends on tracker geometry. Mu3e case: Triplet fit has similar performance as General Broken Lines fit Fast, suitable for filter farm (trigger) Nucl.Instrum.Meth. A844 (2017) 135 Mu3e detector (geometry) 16.5 MeV/c 53 MeV/c Relative momentum resolution 3% 2% 1% 0 (Karimäki) p [MeV/c] 53 Single Helix (Karimäki) Neglect MS Triplets MS fit GBL (General Broken Lines) Fast global track refit with full covariance matrix Equivalent to Kalman filter Alexandr Kozlinskiy 26

27 Mu3e track reconstruction Track reconstruction is based on triplets Fast and similar performance to GBL (Mu3e environment) The same algorithm will be used offline and online Offline implementation also takes into account pixel size and energy loss (minor fitter change) Mu3e readout is essentially continuous: Frame data by combining hits with same (close) timestamps from different system (pixels, fibres, tiles) Reconstruction "frame" of 3 σ t sort hits in time bins Tracking performance: Geant4 simulation of Mu3e detector 50 ns reconstruction frame 10 8 muon stops, Michel decay σ t reconstruction frame Alexandr Kozlinskiy 27

28 Long 6-hit tracks long 6-hit tracks (short + 2 hits) Long 6-hit tracks: Short track + 2 hits Fake rate ~ 9% 1% random combinations Fibre and/or tile hits Reject wrong combinations Alexandr Kozlinskiy 28

29 Acceptance & Efficiency Short tracks (4 hits) Geometrical acceptance: 80% Reconstruction efficiency: 95% Geometrical cuts and χ 2 cuts Long tracks (6 and 8 hits) 80% of short reconstructed as long Geometry (service areas, etc.) χ 2 cuts Efficiency (acceptance) Ends of recurl stations p T acceptance polar angle (outer layer radius) Service areas between stations Alexandr Kozlinskiy 29

30 Timing Time information from fibres/tiles: Suppress fakes Additional vertex constraint Same time at vertex for all tracks 39% fibres Timing information 59% tiles Charge ID: e + or e Mainly for long 8-hit tracks Fibre time difference vs path length 2% no timing Ambiguity for central 8-hit tracks e +? e? Positive time difference: Right charge ID t 1,2 s 1,2 Wrong charge ID Alexandr Kozlinskiy 30

31 Charge ID Charge ID for long 8-hit track: Unknown direction Use fibre time information (t 2 t 1 ) and path length between fibre hits (s 2 s 1 ) Should match particle traveling with speed of light e +? e? Positive time difference: Right charge ID t 1, s 1 t 2, s 2 Wrong charge ID Alexandr Kozlinskiy 31